From Eli in <a href="https://rrtmgp2.slack.com/archives/GCEFZJT9S/p1660314906729399" r

<a href="https://docs.aws.amazon.com/glue/latest/dg/dev-endpoint-tutorial-local-jupyte

<a href="https://stackoverflow.com/questions/20727375/multiprocessing-pool-slower-than

<a href="https://rrtmgp2.slack.com/archives/D942AU7QE/p1664285448033709" rel="nofollow

module has been modified with <a class="commit-link" data-hovercard-type="commit" data

checkpoint at <a class="commit-link" data-hovercard-type="commit" data-hovercard-url="

<a href="https://www.nersc.gov/assets/Uploads/11-IO-Best-Practices-20200616.pdf" rel="

Refinements to Absolute Value and "Parabola" Implementations,about pernak18/g-point-reduction

Comments (13)

pernak18 commented on July 30, 2024

i'm not crazy about the idea of the latter option (retaining 20-30 iterations of information), so i'm thinking about the former -- evaluating delta-CF for starting and modified combinations. so the challenge is to not add more latency to an already-time consuming process.

in this thought process, i'm wondering about the efficiency of a) running this code on cori at NERSC, and b) Python parallelization. I've seen a couple articles suggesting parallelization in Python might not be as fast it seems it would be. the first article also suggests clusters can be problematic with parallelization.

with that in mind, perhaps we can get this software running on a large EC2 instance with a single EBS volume mounted to it? cluster and I/O latency should not be a problem, then. the main question is: does the software run faster on an AWS EC2 instance than it does on a 64-core NERSC supercomputer using an NFS?

from g-point-reduction.

pernak18 commented on July 30, 2024

JupyterLab on EC2 instance

from g-point-reduction.

pernak18 commented on July 30, 2024

i've had some difficulty today running in AWS -- i tried a sandbox in A Cloud Guru but didn't have permissions to start the instance i wanted (i chose a sandbox because i don't have a Project tag for resources yet) and my AWS workspace (but i couldn't build our RRTMGP Garand driver because of gfortran/ifort inconsistencies, and i could only upgrade my WS to 8 vCPUs, anyway) -- so i've tried to run the notebook locally on rd47. i've clocked one iteration locally and one on cori and the results are:

rd47 (8 cores)

Iteration 1
band03_coefficients_LW_g04-05_iter001.nc, Trial: 34, Cost: 92.965532, Delta-Cost: -7.0345
flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 85.5336, 99.5464, 99.7354, 100.0001, 99.3912, 100.0000, 159.2603, 23.3022, 23.4239, 11.9387, 233.1549, 22.1575, 235.7753, 16.4378, 100.0000, 100.0000, 100.0000
35.120123187000004

cori (64 cores)

Iteration 1
band13_coefficients_LW_g15-16_iter001.nc, Trial: 195, Cost: 100.000000, Delta-Cost: 0.0000
flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000, 100.0000
8.277178223

so that's an interesting result. we expect cori iterations to be faster because we are using 8 times as many cores. but it's only about 4 times faster. a couple of notes:

cost results are different because the two notebooks are using two different definitions; the cori run is the newest and uses the absolute value to pick the winner
rd47 is only writing to local disk -- there are no network file shares being accessed
processing times are not the same as wall clock times, and the amount of time i waited was significantly more than what was printed to standard output

from g-point-reduction.

pernak18 commented on July 30, 2024

useful SO post on potentially making parallelization faster

from g-point-reduction.

pernak18 commented on July 30, 2024

Slack conversation between me and eli:

are you not already doing the algorithm for real?

it is pretty close to the real algorithm. the real algorithm might use 0.05 and 0.10 for its calculations, so the actual choice for the winner might be different than the estimate in the choosing algorithm.
10:50
might stretch it up to 0.15. so the choosing algorithm is estimating the gpoints with the most potential for success. (edited)

from g-point-reduction.

pernak18 commented on July 30, 2024

module has been modified with e65ec97. still working on some refinements (aesthetic changes) in the module and notebook.

from g-point-reduction.

pernak18 commented on July 30, 2024

checkpoint at 0fd809f. at this point, we are finding the correct preliminary solution (whatever trial minimized the delta-cost from the previous iteration), and we are at a point where the modified optimization is triggered, but we are not implementing it correctly because we still need to take future iterations into consideration.

my current work is saved in /global/u1/p/pernak18/RRTMGP/test_abs_val on cori. i've done a:

cp LW_cost-optimize_iter093.pickle LW_cost-optimize.pickle

and we can start from here.

from g-point-reduction.

pernak18 commented on July 30, 2024

i see two avenues implementing the modified combination for real. for a given iteration:

setup all trials, and call modCombine() for all trials where their |delta-cost| > 0.1 (or whatever the threshold is), then selecting the winner amongst all of modified and unmodified trials
only apply modCombine() to the trial that was chosen as the preliminary winner and compare its |delta-cost| to the unmodified trials' |delta-cost|

in discussing this with eli, the answer is option 1, keeping in mind that by this point, any trials with |delta-cost| <= 0.1 have already been chosen. the idea now is to apply modCombine to all the remaining trials.

as currently written, modCombine scales the combination weights with unity and 2, but we can omit the former because we just look for a potential zero-crossing, then run a regression to pinpoint the crossing for a given trial. we'll do this for all remaining trials, then pick the winner by running the cost optimization with all the modified trials.

note that with the regression, we have the potential for a lot of ties, because many trials will have zero crossings.

something to remember: i've been focusing a lot on the preliminary winner trial, but it's pretty immaterial at this point. its only purpose was to bring to our attention that we are now modifying the combinations. now we're in it, and we're applying the modification to everything that is left.

from g-point-reduction.

pernak18 commented on July 30, 2024

cori scratch is a Lustre file system. not sure about /home, but i would think it is not not. so we must get some I/O improvement by running the code on scratch. it's faster and bigger. output from showquota on cori:

+-------------+------------+-------------+----------------+------------+-------------+----------------+
| File system | Space used | Space quota | Space used (%) | Inode used | Inode quota | Inode used (%) |
+-------------+------------+-------------+----------------+------------+-------------+----------------+
|        home |   12.28GiB |    40.00GiB |          30.7% |     20.20K |       1.00M |           2.0% |
|    cscratch |    9.39GiB |    20.00TiB |           0.0% |     14.39K |      10.00M |           0.1% |
+-------------+------------+-------------+----------------+------------+-------------+----------------+

but it's purged every 12 weeks. given that cori is being retired in March 2023, this timeline should work for us, but we should move results to Perlmutter scratch when we have something.

from g-point-reduction.

pernak18 commented on July 30, 2024

just did a test on scratch -- copying all the code and reference netCDFs to scratch -- and one iteration still took ~4 minutes, so i guess I/O is not the bottleneck

from g-point-reduction.

pernak18 commented on July 30, 2024

modified cost on single iteration
modification should be its own callable function
are weights updated with modified k-dist? i'm pretty sure they are
setupNextIter() could be updated so that the next iteration only works with band that modified in current iteration -- this would lead to less computation, memory, and i believe disk space consumption
clean up the modified k dist files to save disk space? i'm not entirely sure this is necessary because we're not regenerating them every iteration so there is not a whole lot of redundancy

from g-point-reduction.

pernak18 commented on July 30, 2024

with 5d9d20c, the modified reduction code is ready for further extensive application

from g-point-reduction.

pernak18 commented on July 30, 2024

so i have the results for future reference...94 was the first iteration where we hit the delta-cost threshold, and the

Iteration 94
band16_coefficients_LW_g03-04_iter094.nc, Trial: 145, Cost: 100.558639, Delta-Cost: 0.1196
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.5851, 99.3616, 102.0414, 103.3084, 96.2623, 100.5032, 107.0325, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847

... apply g-point combine modifications...results for 2 different weight scales ("plus" and "2plus"):
band16_coefficients_LW_g03-04_iter094.nc, Trial: 145, Cost: 100.517579, Delta-Cost: 0.5176
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.5145, 99.3939, 102.0424, 103.3090, 96.2623, 100.4943, 107.0325, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847
band16_coefficients_LW_g03-04_iter094.nc, Trial: 145, Cost: 100.474838, Delta-Cost: 0.4748
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.4411, 99.4270, 102.0434, 103.3096, 96.2623, 100.4856, 107.0325, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847

... eventual winner with real flux and cost calcs and the zero-crossing weight scale:
band13_coefficients_LW_g03-04_iter094_regr094.nc, Trial: 131, Cost: 100.438930, Delta-Cost: -0.0001
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.3352, 99.3638, 102.0301, 103.3029, 98.0105, 100.4869, 107.4082, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847

and for iterations 95+:

Iteration 95
95
100%|██████████| 1/1 [00:00<00:00,  8.57it/s]
[-0.32434771  0.13639468  0.32384832  0.56167415  0.66986899]
band04_coefficients_LW_g10-11_iter094_2p095.nc, Trial: 50, Cost: 100.438777, Delta-Cost: -0.0003
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.3784, 99.4155, 99.7234, 105.5774, 96.2623, 100.4869, 107.2385, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847
plus band04
2plus band04
Iteration 96
96
100%|██████████| 3/3 [00:00<00:00,  8.66it/s]
[ 6.25195007  0.57674043  0.10499612  0.33882829 -0.3436686  -0.22453218
  2.41889817 36.39100166 20.23010735  2.10516237  1.37441278 -0.18064123]
band08_coefficients_LW_g08-09_iter094_2p096.nc, Trial: 91, Cost: 100.438609, Delta-Cost: -0.0004
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.4892, 99.1868, 101.7158, 103.2353, 96.2696, 100.4869, 106.9543, 100.0000, 83.2970, 101.0232, 101.3445, 100.4604, 99.9510, 98.5044, 98.2134, 101.6962, 102.0847
plus band08
2plus band08
Iteration 97
97
100%|██████████| 5/5 [00:00<00:00,  5.41it/s]
[-1.79233237 -1.05745992 -1.12991275 -1.09942919  1.23349282  2.88227751
  0.11030006 -1.8756314 ]
band15_coefficients_LW_g02-03_iter094_2p097.nc, Trial: 137, Cost: 100.439686, Delta-Cost: 0.0006
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.4060, 99.4504, 102.0450, 103.3110, 96.2036, 100.4869, 106.6611, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847
plus band15
2plus band15
Iteration 98
98
100%|██████████| 3/3 [00:00<00:00,  8.37it/s]
[-1.12599506 -1.77731129 -0.6801824 ]
band09_coefficients_LW_g09-10_iter094_2p098.nc, Trial: 100, Cost: 100.438325, Delta-Cost: -0.0007
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.2860, 99.4728, 101.8279, 104.3576, 96.3563, 98.1823, 107.0325, 100.0000, 83.2970, 101.2200, 101.4471, 100.4610, 99.9564, 99.2261, 98.1497, 101.5869, 102.0604
plus band09
2plus band09
Iteration 99
99
100%|██████████| 1/1 [00:00<00:00,  6.62it/s]
[ 4.15586056 15.07247601  2.51482418  3.93483337 -0.08418756  0.21878163
  1.00999292  0.6439527 ]
band11_coefficients_LW_g04-05_iter094_2p099.nc, Trial: 110, Cost: 100.439898, Delta-Cost: 0.0009
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.4333, 99.3787, 101.8994, 103.2172, 96.2623, 100.4869, 107.0325, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847
plus band11
2plus band11
Iteration 100
100
100%|██████████| 10/10 [00:01<00:00,  8.11it/s]
[-0.5329217  -1.10818262 -0.67249033 -0.41787364 -0.67113576 -0.57464177
 -0.46644885 -0.72925697 -1.09465086 -0.24856489]
band06_coefficients_LW_g01-02_iter094_2p100.nc, Trial: 67, Cost: 100.438134, Delta-Cost: -0.0009
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.3785, 99.2333, 102.1136, 103.3526, 96.2623, 100.4869, 107.2580, 100.0000, 76.2398, 103.5003, 101.4471, 102.4229, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847
plus band06
2plus band06
Iteration 101
101
100%|██████████| 5/5 [00:00<00:00,  6.72it/s]
[-1.94937854 -2.65382132 -2.48632084  0.73124957 -1.19587914 -1.44394889]
band02_coefficients_LW_g12-13_iter094_2p101.nc, Trial: 25, Cost: 100.436921, Delta-Cost: -0.0021
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.4591, 99.3861, 101.6112, 103.3492, 96.2623, 100.4869, 107.0325, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847
plus band02
2plus band02
 64%|██████▎   | 7/11 [07:02<04:15, 63.87s/it]
Iteration 102

  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:01,  5.68it/s]
 29%|██▊       | 2/7 [00:00<00:00,  5.69it/s]
 43%|████▎     | 3/7 [00:00<00:00,  5.90it/s]
 57%|█████▋    | 4/7 [00:00<00:00,  5.84it/s]
 71%|███████▏  | 5/7 [00:00<00:00,  5.87it/s]
 86%|████████▌ | 6/7 [00:01<00:00,  6.17it/s]
100%|██████████| 7/7 [00:01<00:00,  4.57it/s]
band07_coefficients_LW_g09-10_iter094_2p102.nc, Trial: 80, Cost: 100.436866, Delta-Cost: -0.0022
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.3803, 99.4596, 103.1619, 102.1643, 96.2623, 100.4869, 107.0422, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847
plus band07
2plus band07
 73%|███████▎  | 8/11 [09:00<04:03, 81.07s/it]
Iteration 103

  0%|          | 0/9 [00:00<?, ?it/s]
 11%|█         | 1/9 [00:00<00:01,  6.81it/s]
 22%|██▏       | 2/9 [00:00<00:01,  6.04it/s]
 33%|███▎      | 3/9 [00:00<00:00,  6.45it/s]
 44%|████▍     | 4/9 [00:00<00:00,  6.26it/s]
 56%|█████▌    | 5/9 [00:00<00:00,  6.35it/s]
 67%|██████▋   | 6/9 [00:00<00:00,  6.42it/s]
 78%|███████▊  | 7/9 [00:01<00:00,  6.46it/s]
 89%|████████▉ | 8/9 [00:01<00:00,  6.55it/s]
100%|██████████| 9/9 [00:01<00:00,  6.45it/s]
band12_coefficients_LW_g04-05_iter094_2p103.nc, Trial: 117, Cost: 100.441626, Delta-Cost: 0.0026
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.4007, 99.5034, 102.0596, 103.3359, 96.2623, 100.4869, 106.6022, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847
plus band12
2plus band12
 82%|████████▏ | 9/11 [10:38<02:52, 86.33s/it]
Iteration 104

  0%|          | 0/7 [00:00<?, ?it/s]
 14%|█▍        | 1/7 [00:00<00:00,  6.32it/s]
 29%|██▊       | 2/7 [00:00<00:00,  6.54it/s]
 43%|████▎     | 3/7 [00:00<00:00,  6.77it/s]
 57%|█████▋    | 4/7 [00:00<00:00,  7.10it/s]
 71%|███████▏  | 5/7 [00:00<00:00,  6.07it/s]
 86%|████████▌ | 6/7 [00:00<00:00,  6.61it/s]
100%|██████████| 7/7 [00:01<00:00,  6.43it/s]
band03_coefficients_LW_g12-13_iter094_2p104.nc, Trial: 39, Cost: 100.442573, Delta-Cost: 0.0035
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.3809, 99.4383, 102.0138, 103.3627, 96.2623, 100.4869, 107.0459, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847
plus band03
2plus band03
 91%|█████████ | 10/11 [11:55<01:23, 83.64s/it]
Iteration 105

  0%|          | 0/6 [00:00<?, ?it/s]
 17%|█▋        | 1/6 [00:00<00:00,  5.18it/s]
 33%|███▎      | 2/6 [00:00<00:00,  5.53it/s]
 50%|█████     | 3/6 [00:00<00:00,  5.62it/s]
 67%|██████▋   | 4/6 [00:00<00:00,  6.05it/s]
 83%|████████▎ | 5/6 [00:00<00:00,  6.06it/s]
100%|██████████| 6/6 [00:01<00:00,  5.71it/s]
band10_coefficients_LW_g02-03_iter094_2p105.nc, Trial: 97, Cost: 100.435146, Delta-Cost: -0.0039
	flux_net, band_flux_net, heating_rate, heating_rate_7, flux_net_forcing_5, flux_net_forcing_6, flux_net_forcing_7, flux_net_forcing_9, flux_net_forcing_10, flux_net_forcing_11, flux_net_forcing_12, flux_net_forcing_13, flux_net_forcing_14, flux_net_forcing_15, flux_net_forcing_16, flux_net_forcing_17, flux_net_forcing_18 = 99.3811, 99.5521, 102.0036, 103.2767, 96.2623, 100.4869, 107.0325, 100.0000, 83.2970, 101.2200, 101.4471, 100.4604, 99.9562, 99.2261, 98.1497, 101.5854, 102.0847
plus band10
2plus band10
100%|██████████| 11/11 [13:31<00:00, 73.76s/it]

from g-point-reduction.

Refinements to Absolute Value and "Parabola" Implementations about g-point-reduction HOT 13 CLOSED

Comments (13)

Related Issues (20)

Recommend Projects

React

Vue.js

Typescript

TensorFlow

Django

Laravel

D3

Recommend Topics

javascript

web

server

Machine learning

Visualization

Game

Recommend Org

Facebook

Microsoft

Google

Alibaba

D3

Tencent